Learning From Drivers Patterns of Evasive Maneuvers in Case of Emergency Using Dynamic-Time-Warping-Based Clustering.

How to ensure safe takeovers in Level 3 automated driving has become a great concern. Among all the studies, the research question “What if drivers are not ready or fail to take over within required time?” seems to have been less emphasized. Therefore, the main objective of this research is to learn from drivers patterns of evasive maneuvers in emergent cases using dynamic-time-warping (DTW)-based k-means clustering. Forty-eight participants were recruited for a driving simulator study, and application of the algorithm resulted in three patterns of evasive maneuvers irrespective of scenarios, with Type I and III being the least and the most cautious ones. Specifically, drivers with Type I maneuvers preferred to overtake the vehicle ahead at higher speeds, lower accelerations, and quicker lane change maneuvers. Drivers with Type II maneuvers would slow down the vehicle before changing lanes, and then started to accelerate and perform the lane change. Drivers with Type III maneuvers tended to decelerate the vehicle to a low speed and kept at that speed for some time before they started to accelerate and finish the lane change. This helped to understand and design evasive maneuvers during automated driving and paved the way for the next research.